I'm trying to summarize the frequency domain information of an electrical signal as a feature engineering step for a machine learning algorithm. My hypothesis is that an anomaly will appear in the power signal as harmonics and spikes that can be detected in the frequency domain.

I've used MFCC in the past for audio features, but since this signal belongs to a different domain I think that Mel Scale would not be a good choice for this task.

Are there any alternatives to MFCCs on different scales? I guess even a linear scale would be better as a naive first approach.

Or perhaps there are other transformations/methods better suited to summarize the frequency content of my signal.

  • $\begingroup$ There is nothing that prevents you from using a different scale for placing the triangular filtebank or even using other filter shapes such as rectangular. $\endgroup$ – jojek Mar 8 '18 at 13:08
  • $\begingroup$ @jojek are you aware of any implementation of this? $\endgroup$ – jimijazz Mar 8 '18 at 13:11
  • 1
    $\begingroup$ Well, you might actually get away with doing the DCT of the plain power spectrum in log domain. $\endgroup$ – jojek Mar 8 '18 at 18:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.